4D Visual Grounding and Tracking on Instruct-4D macro-averaged across EgoWL and AlloWL 1.0
62.68TGATop14DTrack-Qwen3.5-9B
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| 4DTrack-Qwen3.5-9BBackbone=Qwen3.5-9B, Framework=4DTrack2026.06 | 62.68 | 51.93 | 55.18 | 85.57 | 3.67 | 58.27 | |
| 4DTrack-Qwen2.5-VL-7B-InstructBackbone=Qwen2.5-VL-7B-Instruct, Framework=4DTrack2026.06 | 50.37 | 48.91 | 47.61 | 74.69 | 4.71 | 47.24 | |
| UVLTrackCategory=VLT Baselines2026.06 | 43.06 | 17.43 | 21.84 | 55.76 | 8.33 | 25.58 | |
| DUTrackCategory=VLT Baselines2026.06 | 42.48 | 18.03 | 22.17 | 56.38 | 8.05 | 25.66 | |
| GLADCategory=VLT Baselines2026.06 | 42.18 | 19.65 | 22.08 | 52.54 | 8.2 | 26.08 | |
| JointNLTCategory=VLT Baselines2026.06 | 37.04 | 15.17 | 18.89 | 48.79 | 8.82 | 26.08 | |
| 4DTrack-Mistral-7B-v0.2Backbone=Mistral-7B-v0.2, Framework=4DTrack2026.06 | 31.53 | 28.04 | 30.11 | 67.84 | 8.05 | 29.34 | |
| 4DTrack-Llama-3-8B-InstructBackbone=Llama-3-8B-Instruct, Framework=4DTrack2026.06 | 25.16 | 20.66 | 21.45 | 64.04 | 11.93 | 21.39 | |
| Qwen3.5-9BCategory=Open-Source Models, Interface=shared 4DVLT interface2026.06 | 14.12 | 10.13 | 13.99 | 55.9 | 13.71 | 11.38 | |
| Qwen2.5-VL-7B-InstructCategory=Open-Source Models, Interface=shared 4DVLT interface2026.06 | 11.03 | 22.74 | 15.12 | 51.1 | 13.78 | 14.79 | |
| Mistral-7B-v0.2Category=Open-Source Models, Interface=shared 4DVLT interface2026.06 | 7.51 | 20.98 | 14.22 | 51.69 | 13.8 | 12.36 | |
| Llama-3-8B-InstructCategory=Open-Source Models, Interface=shared 4DVLT interface2026.06 | 6.44 | 19.28 | 13.61 | 50.02 | 13.6 | 12.72 |